improvable AI behaviour

[Rokko]

Ok, we all know that most Combat Mission scenarios tend to place you into the role of the attacker and that the reason for this is the AI not being very smart when it comes to attacking.

The problem is, of course, that this takes away many possible scenario ideas. We are actually reduced essentially to German side scenarios recreating instances where the Germans were counter-attacking in Normandy (like the Mortain counter-offensive). Fun things like flexible and in-depth bocage defense as player can actually only happen in Multiplayer.

I've created a small testing map to see, what it is, that makes the AI incapable as attackers.

Now I have not the slighest idea about AI programming and I don't know what can be actually be realised and what not.

Anyways. In the test scenario a German AI Grenadier platoon minus its Panzerschrecks is tasked with assaulting over open ground towards a bunch of houses, with two lenghty wall sections providing cover.

This is the situation. I've used terrain objectives to show where the AI orders for the platoon are placed. The first 4 squares are the setup areas.

My intention was for the platoon to advance along the 4 indicated approaches as highlighted.

Now, a human player would probably place each squad in one of the squares, with the HQ somewhere in the middle to maintain optimal C2.

The AI doen't take this into consideration, though. Sometimes it does place the HQ in the middle, sometimes it doesn't, its totally random. In this case the HQ is on the far left. Sometimes it even clutches two squads into one square, next pic shows one of those instances.

To reach the first stage a human player would naturally send each squad to the nearest section of the wall, straight ahead of them, to maximize speed and minimize exposure to the enemy.

This is where AI madness starts...

[Rokko]

The next AI order is "Assault", so the the AI will use the assault movement and keep one or two squads back to provide fire support.

But again no rational behaviour, as the AI decides to let the two advancing squads cross their paths.

The next AI order is "Assault", so the the AI will use the assault movement and keep one or two squads back to provide fire support.

But again no rational behaviour, as the AI decides to let the two advancing squads cross their paths.

The last squad and the HQ then did something smart, they moved to next wall using the intended line of approach, instead of making a stupid diagonal movement from the outer left to the outer right of the 2nd stage.

But again, I'm afraid this is random.

My conclusions are from this test that the AI doesn't evaluate how to use short ways to minimize exposure and time. Also, it doesn't seem to "think ahead". It seems to only plan from one AI order to the next, instead of "seeing" its final goal.

Some things are desperatly needed, like sub-AI group control or more AI groups instead. In bigger scenarios its not possible to have every platoon be a single AI group. Instead you'd have a whole company running around like this.

Also triggers would be really nice, also for timing AI artillery. But pretty basic things could really improve the way the scenario designer can control things, like: "IF group X has suffered Y% casualties THEN move (retreat) to Zone Z".

I think, someone mentioned they are going to improve the coordination between AI groups, which would definately great, too.

Now with that being said, I wanted to mention, that, while the scenario AI is subject to some criticism, the TacAI in this game is really superb, I really adore, despite some minor issues like firing small arms at tanks which is obviously going to be fixed sooner or later.

[Chris Ferrous]

Nice little test, Rokko.

Your assessment fits with what I've seen also in trying to get the AI to move logically although I've not seen it behave in quite suchan extreme way.

Can I suggest that you amend your test map so that each 'phase line' is a continuous strip rather than isolated spots, and then run the test again to see what happens?

I think that would be very helpful.

Also, when you've assessed it for infantry would you do it for tanks? At the moment I have a feeling that under AI control tanks will turn to face their intended next position almost as soon as they arrive at their first spot. I've been placing a 'beyond scenario time limit' location such that the tanks will face forward.

Also I can see no way of getting tanks to retreat without turning their backsides towards the enemy. :confused:

[Erik Springelkamp]

I have done similar tests in CMSF, and also when the destination areas are continuous, the AI moves a lot in diagonal lines across the front.

They also arrive piecemeal without sufficient mutual support.

With tanks you can often take out one after the other attacking tank as they come into view.

[Rokko]

I've tried using continuous strips already, it didn't change much, the AI did just have greater choice where to move its units to (nonetheless in a crazy way).

In another test I have also added an HMG team to the platoon and was pleased to see that the AI kept it in the back during the assault order.

I didn't try AI scripting and tanks out alot, but it seems that Assault and Max Assault don't work for vehicles.

For your problem you might try to add another order right next to the last and in the direction where you want the tank(s) to face, but with the time limit of the order you want them approach next.

This way they might face in the wanted direction, wait in their spot as long as you want them, then move a little (one tile further) and immediatly after that to the next order.

[SteveP]

I have been studying the AI scripting feature quite a bit since getting CMBN, and there is a lot of depth to it, but also some notable handicaps, which are hard to work around sometimes. Still it can produce a more challenging AI than was the case in CM1, especially when the AI is attacking. However ...

The criss-crossing movement is a puzzle. I assume this behavior is an artifact of the path-finding program in the OpAI. And I assume the path-finding is done unit by unit (not formation by formation). But I can't think of any reason why the OpAI would choose a longer path over a shorter one, assuming cover/concealment are equal. It would be great if BFC would comment on this at some point.

[Childress]

You get the impression that the number of wargames that can provide a challenging AI opponent on the offence is pretty rare. Black Swan rare. Can't think of any offhand. Although some claim 'Battles from the Bulge' isn't too shabby.